To conduct regional simulations of carbon accumulation for both uplands and peatlands, we first parameterized the P-TEM for representative ecosystems in Alaska. Second, we organized the regional vegetation and peatland distribution data, spatial basal age data for all peatland grid cells based on site-level soil core data, and climate data for each period during the Holocene. Finally, we conducted the regional simulations and sensitivity analysis.

In P-TEM (Wang et al., 2016), peatland soil organic C (SOC) accumulation is determined by the difference between NPP and aerobic and anaerobic decomposition. Peatlands accumulate C where NPP is greater than decomposition, resulting in positive net ecosystem production (NEP): NEP=NPP-RH-RCH4-RCWM-RCM-RCOM. P-TEM was developed based on the Terrestrial Ecosystem Model (TEM) at a monthly step (Zhuang et al., 2003, 2015). It explicitly considers the process of aerobic decomposition (RH) related to the variability of water-table depth; net methane emission after methane oxidation (RCH4); CO2 emission due to methane oxidation (RCWM) (Zhuang et al., 2015); CO2 release accompanied by the methanogenesis (RCM) (Tang et al., 2010; Conrad, 1999); and CO2 release from other anaerobic processes (RCOM, e.g., fermentation, terminal electron acceptor (TEA) reduction) (Keller and Bridgham, 2007; Keller and Takagi, 2013). For upland soils, we only considered the heterotrophic respiration under aerobic conditions (Raich et al., 1991). For a detailed model description, see Wang et al. (2016).

We modeled peatland soils as a two-layer system for a hydrological module (HM) while keeping the three-layer system for upland soils (Zhuang et al., 2002). The soil layers above the lowest water-table position are divided into (1) a moss (or litter) organic layer (0–10 cm) and (2) a humic organic layer (10–30 cm) (Wang et al., 2016). Based on the total amount of water content within those two unsaturated layers, the actual water-table depth (WTD) is estimated. The water content at each 1 cm above the water table can then be determined after solving the water balance equations (Zhuang et al., 2004).

In the STM module, the soil vertical profile is divided into four layers: (1) snowpack in winter, (2) a moss (or litter) organic layer, and (3) upper and (4) lower humic organic soil (Wang et al., 2016). Each of these soil layers is characterized by a distinct soil thermal conductivity and heat capacity. We used the observed water content to drive the STM (Zhuang et al., 2001).

The methane dynamics module (MDM) (Zhuang et al., 2004) considers the processes of methanogenesis, methanotrophy, and the transportation pathways, including (1) diffusion through the soil profile, (2) plant-aided transportation and (3) ebullition. The soil temperatures calculated from STM, after interpolation into 1 cm sub-layers, are input to the MDM. The water-table depth and soil water content in the unsaturated zone for methane production and emission are obtained from HM, and NPP is calculated from the CNDM. Soil-water pH is prescribed from observed data and the root distribution determines the redox potential (Zhuang et al., 2004).

We have parameterized the key parameters of the individual modules, including HM, STM, and MDM in Wang et al. (2016). The parameters in CNDM for upland soils and vegetation have been optimized in the previous studies (Zhuang et al 2002, 2003; Tang and Zhuang, 2008). Here we parameterized P-TEM for peatland ecosystems using data from a moderately rich Sphagnum spp. open fen (APEXCON) and a Sphagnum–black spruce (Picea mariana) bog (APEXPER) (Table 1). Both are located in the Alaskan Peatland Experiment (APEX) study area, where Picea mariana is the only tree species above breast height in APEXPER. Three water-table position manipulations were established in APEX, including control, lowered, and raised water-table plots (Chivers et al., 2009; Turetsky et al., 2008; Kane et al., 2010; Churchill et al., 2011). There were also several internal collapse scars that formed with thaw of surface permafrost, including non, old, and new collapse plots. APEXCON represents the control manipulation and APEXPER represents the non-collapse plot. The annual NPP and aboveground biomass at both sites were measured in 2009. There were no belowground observations at APEX, however, at a Canadian peatland, Mer Bleue, which includes Sphagnum spp. dominated bog (dominated by shrubs and Sphagnum) and pool fen (dominated by sedges and herbs and Sphagnum). The belowground biomass was also observed at Suurisuo mire complex, southern Finland, a sedge fen site dominated by Carex rostrate. We used the ratio (70 %) of belowground biomass to total biomass from these two study sites to calculate the missing belowground biomass values at APEXCON and APEXPER (Table 2). We conducted 100 000 Monte Carlo ensemble simulations to calibrate the model for each site using a Bayesian approach, and parameter values with the modes in their posterior distributions were selected (Tang and Zhuang, 2008, 2009).

The Alaskan C stock was simulated through the Holocene driven with vegetation data reconstructed for four time periods, including a time period encompassing a millennial-scale warming event during the last deglaciation known as the Bølling–Allerød at 15–11 ka (1 ka = 1000 cal yr before present), Holocene Thermal Maximum (HTM) during the early Holocene at 11–10 and 10–9 ka, and the mid (9–5 ka) and late Holocene (5 ka–1900 AD) (He et al., 2014). We used the modern vegetation distribution for the simulation during the period 1900–2000 AD (Fig. 2). We assumed that the vegetation distribution remained static within each corresponding time period. Upland ecosystems were classified into boreal deciduous broadleaf forest, boreal evergreen needleleaf and mixed forest, alpine tundra, wet tundra, and barren lands (Table 3). By using the same vegetation distribution map, we reclassified the upland ecosystems into two peatland types, including Sphagnum spp. poor fens (SP) dominated by tundra and Sphagnum spp.–black spruce (Picea mariana) bog/peatland (SBP) dominated by forest ecosystems (Table 3).

The upland and peatland ecosystem distribution for each grid cell was determined using the wetland inundation data extracted from the NASA/GISS global natural wetland dataset (Matthews and Fung, 1987). The resolution was resampled to 0.5∘×0.5∘ from 1∘×1∘. Given the same topography of Alaska during the Holocene, we assumed that the wetland distribution remained the same throughout the Holocene. The inundation fraction was assumed to be the same within each grid through time and the land grids not covered by peatland were treated as uplands. We calculated the total area of modern Alaskan peatlands to be 302 410 km2, which was within the range from 132 000 km2 (Bridgham et al., 2006) to 596 000 km2 (Kivinen and Pakarinen, 1981). The soil water pH data were extracted from Carter and Scholes (2000), and the elevation data were derived from Zhuang et al. (2007).

Our regional simulations considered the effects of basal ages on carbon accumulation. To obtain the spatially explicit basal age data for all peatland grid cells, we first categorized the observed basal ages of peat samples from Gorham et al. (2012) into different time periods, including 15–11, 11–10, 10–9, and 9 ka–19th (Fig. 2). For each time period, the areas dominated by different vegetation types were assigned with varying peatland basal ages. To do that, we examined the association of peat basal ages and vegetation types from peat-core data. For instance, we found that peatland initiations during 15–11 ka occurred in the regions that were dominated by alpine tundra at the southern, northwestern, and southeastern coasts. We thus assign the different peatland basal ages for the grid cells according to their vegetation types for each time slice (Table 4).

Climate data were bias-corrected from ECBilt-CLIO model output (Timm and Timmermann, 2007) to minimize the difference from CRU data (He et al., 2014). Climate fields include monthly precipitation, monthly air temperature, monthly net incoming solar radiation, and monthly vapor pressure at a resolution of 2.5∘×2.5∘. We used the same time-dependent forcing atmospheric carbon dioxide concentration data for model input as were used in ECBilt-CLIO transient simulations from the Taylor Dome (Timm and Timmermann, 2007). The historical climate data used for the simulation through the 20th century were monthly CRU2.0 data (Mitchell et al., 2004).

Simulations for pixels located on the Kenai Peninsula from 15 to 5 ka were first conducted with the parameterized model. The peat-core data from four peatlands on the Kenai Peninsula, Alaska (Jones and Yu, 2010; Yu et al., 2010) (Table 5; also see Table 3 in Wang et al., 2016) were used to compare with the simulations. The observed data include the peat depth, bulk density of both organic and inorganic matters at 1 cm interval, and age determinations. The simulated C accumulation rates represent the actual (“true”) rates at different times in the past. However, the calculated accumulation rates from peat cores are considered “apparent” accumulation rates, as peat would continue to decompose since the time of formation until the present when the measurement was made (Yu, 2012). To facilitate comparison between simulated and observed accumulation rates, we converted the simulated “true” accumulation rates to “apparent” rates, following the approach by Spahni et al. (2013). That is, we summed the annual net C accumulation over each 500-year interval and deducted the total amount of C decomposition from that time period, then dividing by 500 years.

Second, we conducted a transient regional simulation driven with monthly climatic data (Fig. 3) from 15 ka to 2000 AD. The simulation was conducted assuming all grid cells were taken up by upland ecosystems to get the upland soil C spatial distributions during different time periods. We then conducted the second simulation assuming all grid cells were dominated by peatland ecosystems following Table 3 to obtain the distributions of peat SOC accumulation. Finally, we used the inundation fraction map to extract both uplands and peatlands and estimated the corresponding SOC stocks within each grid, which were then summed up to represent the Alaskan SOC stock. We also used the observed mean C content of 46.8 % in peat mass and bulk density of 166 ± 76 kg m-3 in Alaska (Loisel et al., 2014) to estimate peat depth distribution from the simulated peat SOC density (kg C m-2).

Third, we conducted a series of extra simulations to further examine how uncertain climates and vegetation distribution affect our results. We used the original forcing data as the standard scenario and the warmer (monthly temperature +5∘) and cooler (-5∘) as other two scenarios while keeping the rest of the forcing data unchanged. Similarly, we used the original forcing data as the standard scenario and the wetter (monthly precipitation +10 mm) and drier (-10 mm) to test the effect from precipitation. To further study whether vegetation distribution has stronger effects on SOC accumulation than climate in Alaska, we simply replaced SBP with SP and replaced the upland forests with tundra at the beginning of 15 ka. We then conducted the simulation under “warmer” and “wetter” conditions simultaneously as described before while keeping the vegetation distribution unchanged.

Our paleosimulations showed a large peak of peat C accumulation rates at 11–9 ka during the HTM (Fig. 4). The simulated “true” and “apparent” rates captured this primary feature in peat-core data at almost all sites (Jones and Yu, 2010; see Wang et al., 2016, Table 3 for site details). We simulated an average of the peat SOC “apparent” accumulation rate of 11.4 g C m-2 yr-1 from 15 to 5 ka, which was slightly higher than the observations at four sites (10.45 g C m-2 yr-1). The simulated rate during the HTM was 26.5 g C m-2 yr-1, up to 5 times higher than the rest of the Holocene (5.04 g C m-2 yr-1). This corresponded to the observed average rate of 20 C m-2 yr-1 from 11.5 to 8.6 ka, which is 4 times higher than 5 C m-2 yr-1 over the rest of the Holocene.

Model simulations showed an overall low vegetation C before the HTM (15–11 ka) (Fig. 5a) parallel to the relatively low annual and long-term NPP (Fig. 5b and c). The lowest amount of C (∼ 0.8 kg C m-2) was stored in Sphagnum-dominated peatland. Sphagnum–black spruce peatland also had low vegetation C density (∼ 1 kg C m-2). Upland vegetation showed a generally higher C storage, of which boreal evergreen needleleaf forest ranked first (∼ 2 kg C m-2). Highest NPP accompanied by highest vegetation carbon appeared during the HTM (11–9 ka) (Fig. 5a and b). Lower annual C uptake along with lower C was found during the mid and late Holocene (9 ka–19th), where peatland ecosystems exhibited the most obvious drops (Fig. 5a and b).

In general, vegetation held about 2 Pg C before the HTM (Fig. 6). Upland tundra ecosystems accounted for the highest amount of C. During the HTM, boreal evergreen needleleaf forest reached its peak and had an overwhelming proportion over total C. Similarly, a peak of total vegetation C appeared at the same time, averaging around 4.3 Pg C. A large decrease occurred at the mid-Holocene and a slight decline continued till the late Holocene. We estimated a total of 2.9 Pg C stored in modern Alaskan vegetation, with 0.4 Pg in peatlands and 2.5 Pg in non-peatlands. The uncertainties during the model calibration (Table 2) resulted in 0.3–0.6 and 2.2–3.1 Pg C in peatlands (see Wang et al., 2016, for model parameters) and non-peatland vegetation (see Tang and Zhuang, 2008, for uncertainty analyses for upland vegetation), respectively. Our estimation of 2.5–3.7 Pg C stored in the Alaskan vegetation was lower than the previous estimate of 5 Pg (Balshi et al., 2007; McGuire et al., 2009), presumably due to the prior ranges of model parameters used from Tang and Zhuang (2008). Our overestimation of peatland area may also lead to a reduction of Alaskan non-peatland area.

Carbon storage in Alaskan non-peatland soils varied spatially (Fig. 7). Moist tundra had the highest SOC density (12–25 kg C m-2), followed by deciduous broadleaf forest (8–13 kg C m-2) and evergreen needleleaf forest (3–8 kg C m-2) through all time slices between 15 ka and 2000 AD. Dramatic changes in vegetation types have occurred in Alaska during different periods (Fig. 2). Before the HTM (15–11 ka), the terrestrial ecosystem was dominated by tundra. The northwestern coast and the eastern interior were covered by moist tundra. Southwestern Alaska and the interior south of the Brooks Range were dominated by alpine tundra (Fig. 2a). The basal ages of peat samples from Gorham et al. (2012) suggested that peatlands were likely to form from the (alpine) tundra ecosystems, although patches of boreal deciduous broadleaf forest and boreal evergreen needleleaf and mixed forest appeared at the north of the Alaska Range. Initially, only Sphagnum open peatland (SP) existed, with less C (< 10 kg C m-2) sequestrated in the southeastern Brooks Range in comparison with southwestern and northwestern coastal parts (> 15 kg C m-2) (Fig. 8a). Approximately 4.5×105 km2 area was covered by peatlands at the beginning of the HTM (∼ 11 ka) (Fig. 9). During the HTM (11–9 ka), boreal deciduous broadleaf and boreal evergreen needleleaf and mixed forests expanded (Fig. 8b and c). Coastal tundra (moist wet tundra) covered north of the Brooks Range between 11 and 10 ka, where SP continued its expansion (Fig. 8b). Sphagnum–black spruce forested peatland began forming in the southwestern coast and eastern interior regions, with a rapid increase in total peatland area to about 13×105 km2 (Fig. 9). At 10–9 ka, boreal deciduous forest expanded to north of the Brooks Range, making forest the dominant biome in Alaska (Fig. 2c). Prevailing forest ecosystems indicated a large expansion of peatland, with SBP covering the interior of Alaska (Fig. 8c). During the mid-Holocene (9–5 ka), the terrestrial landscape generally resembled present-day ecosystems (Bigelow et al., 2003). Boreal evergreen needleleaf and mixed forest prevailed in southern and interior Alaska, with tundra returning to north of the Brooks Range and western Alaska (Fig. 2d and e). Although SP kept forming towards the west, some areas dominated by SBP in interior Alaska ceased accumulating C (Fig. 8d). At 5k–19th, almost all the peatlands have formed, with some interior regions exhibiting a C loss (Fig. 8e). C accumulation increased again in the last century, averaging about 20 kg C m-2 kyr-1 (Fig. 8f). We found that the distribution of SOC densities of both upland and peatland varied greatly depending on the vegetation distribution within each time slice, indicating that vegetation composition might be a major factor controlling regional C dynamics.

An average peat SOC “apparent” accumulation rate of 13 g C m-2 yr-1 (2.3 Tg C yr-1 for the whole of Alaska) was estimated from 15 ka to 2000 AD (Fig. 10), lower than 18.6 g C m-2 yr-1 as estimated from peat cores for northern peatlands (Yu et al., 2010), and slightly higher than the observed rate of 13.2 g C m-2 yr-1 from four peatlands in Alaska (Jones and Yu, 2010). A simulated peak occurred during the HTM with the rate 29.1 g C m-2 yr-1 (5.1 Tg C yr-1), which was slightly higher than the observed 25 g C m-2 yr-1 for northern peatlands and ∼ 20 g C m-2 yr-1 for Alaska (Yu et al., 2010). It was almost 4 times higher than the rate 6.9 g C m-2 yr-1 (1.4 Tg C yr-1) over the rest of the Holocene, which corresponded to the peat-core-based observations of ∼ 5 g C m-2 yr-1. The mid and late Holocene showed much slower C accumulation at a rate of approximately 5-fold lower than during the HTM. This corresponded to the observation of a 6-fold decrease in the rate of new peatland formation after 8.6 ka (Jones and Yu, 2010). The C accumulation rates increased abruptly to 39.2 g C m-2 yr-1 during the last century, within the field-measured average apparent rate range of 20–50 g C m-2 yr-1 over the last 2000 years (Yu et al., 2010).

The SOC stock of northern peatlands has been estimated in many studies, ranging from 210 to 621 Pg (Oechel, 1989; Gorham, 1991; Armentano and Menges, 1986; Turunen et al., 2002; Yu et al., 2010; see Yu, 2012, for a review). Assuming Alaskan peatlands were representative of northern peatlands and using the area of Alaskan peatlands (0.45×106 km2; Kivinen and Pakarinen, 1981) divided by the total area of northern peatlands (∼ 4×106 km2; Maltby and Immirzi, 1993), we estimated a SOC stock of 23.6–69.9 Pg C for Alaskan peatlands. Our model estimated 27–48 Pg C (23.9 Pg C in SP and 13.5 Pg C in SBP) had been accumulated from 15 ka to 2000 AD (Fig. 11), due to uncertain parameters (Table 2; see Wang et al., 2016, for model parameters). The uncertainty can also have resulted from peat basal age distributions and the estimation of total peatland area using modern inundation data as discussed above. By incorporating the observed basal age distribution to determine the expansion of peatland through time, we estimated that approximately 68 % of Alaskan peatlands had formed by the end of the HTM, similar to the estimation from observed basal peat ages that 75 % peatlands had formed by 8.6 ka (Jones and Yu, 2010).

The northern circumpolar soils were estimated to cover approximately 18.78×106 km2 (Tarnocai et al., 2009). The non-peatland soil C stock was estimated to be in the range of 150–191 Pg C for boreal forests (Apps et al., 1993; Jobbágy and Jackson, 2000) and 60–144 Pg C for tundra in the 0–100 cm depth (Apps et al., 1993; Gilmanov and Oechel, 1995; Oechel et al., 1993); 1.24×106 km2 non-peatland area was estimated from the total land area of Alaska (1.69×106 km2). Therefore, Alaska non-peatland soil contained 17–27 Pg C by using the ratio of Alaskan over northern non-peatland. In comparison, we modeled 9–15 Pg C (within 1 m depth), depending on the prior ranges of model parameters from Tang and Zhuang (2008).

The simulated modern SOC distribution (Fig. 12c) was largely consistent with the study of Hugelius et al. (2014) (see Fig. 3 in the paper). The model captured the SOC density on the northern and southwestern coasts of Alaska, with most grids > 40 kg C m-2 on average. Those regions also showed high SOC density (> 75 kg C m-2), which was also exhibited in our result. The eastern part and the western coast had the lowest SOC densities, corresponding to the simulation result that most grids had SOC values between 20 and 40 kg C m-2. We estimated an average peat depth of 1.9 ± 0.8 m considering the uncertainties within dry bulk densities. It was similar to the observed mean depth of 2.29 m for Alaskan peatlands (Gorham et al., 1991, 2012). Our estimates (Fig. 12d) showed a relatively high correlation with the 64 observed peat samples, especially with higher depths (Fig. 13) (R2=0.45). The large intercept of the regression line (101 cm) suggested that the model might have not performed well in estimating the grids with low peat depths (< 50 cm). The peat characteristics (e.g., bulk density) from location to location may differ greatly, even within the same small region. Thus, it is difficult to capture the observed variations of peat depths, as we used the averaged bulk density of the whole of Alaska.

The simulated climate by the ECBilt-CLIO model showed that among the six time periods, the coolest temperature appeared at 15–11 ka, followed by the mid and late Holocene (5 ka–1900 AD). Those two periods were also generally dry (Fig. 3f). The former represented colder and drier climates before the onset of the Holocene and the HTM (Barber and Finney, 2000; Edwards et al., 2001). The latter represented post-HTM neoglacial cooling, which has caused permafrost aggradation across northern high latitudes (Oksanen et al., 2001; Zoltai, 1995).

Despite the relatively large inter-annual NPP variation that resulted from the annual fluctuations of the climate forcing (Fig. 5b), the long-term NPP, vegetation C density and storage were highest during the HTM (Fig. 5a and c). Annual C accumulation rates also reached their peak (Figs. 5–11). The long-term variation of NPP has a similar pattern of the climate (see Fig. 3 for climate variables), where higher NPP, along with higher vegetation C, coincided with warmer temperatures and enhanced precipitation during the HTM compared to other time periods. ECBilt-CLIO simulated a warmest summer and a prolonged growing season, leading to a stronger seasonality of temperature during the HTM (Kaufman et al., 2004, 2016), in line with the orbitally induced maximum summer insolation (Berger and Loutre, 1991; Renssen et al., 2009). The coincidence between the highest vegetation C uptake and SOC accumulation rates and the warmest summer and the wetter-than-before conditions indicated a strong link between those climate variables and C dynamics in Alaska. Enhanced climate seasonality characterized by warmer summer, enhanced summer precipitation and possibly earlier snowmelt during the HTM accelerated the photosynthesis and subsequently increased NPP (Tucker et al., 2001; Kimball et al., 2004; Linderholm, 2006). As shown in our sensitivity test, annual NPP was increased by 40 and 20 g C m-2 yr-1 under the warmer and wetter scenarios, respectively (Fig. 14a, b). Meanwhile, warmer condition could positively affect the SOC decomposition (Nobrega et al., 2007). However, it could be offset to a certain extent via the hydrological effect, as higher precipitation could raise the water-table position, allowing less space for aerobic heterotrophic respiration. Our sensitivity test results indicated that warmer and wetter conditions could lead to an increase in decomposition of up to 35 and 15 g C m-2 yr-1, respectively (Fig. 14c, d). We did not find a decrease in total heterotrophic respiration throughout Alaska from the higher precipitation. It was presumably due to a much larger area of upland soils (1.3×106 km2) than peatland soils (0.26×106 km2), as higher precipitation would cause higher aerobic respiration in the unsaturated zone of upland soils and consequently stimulate the SOC decomposition. The relatively low NPP and vegetation C density, along with the lower total soil C stocks, were consistent with the unfavorable cool and dry climate conditions at 15–11 ka and during the mid and late Holocene. Statistical analysis indicated that temperature had the most significant effect on peat SOC accumulation rate, followed by the seasonality of NIRR (Wang et al., 2016). The seasonality of temperature, the interaction of temperature and precipitation, and precipitation alone also showed significance. The strong link between climate factors and C dynamics may explain the lower SOC accumulation during the neoglacier cooling period (Marcott et al., 2013; Vitt et al., 2000; Peteet et al., 1998; Yu et al., 2010). The rapid peat SOC accumulation during the 20th century under warming and wetter climate may suggest a continuous C sink in this century, as concluded in Spahni et al. (2013). However, the rising temperature in the future may have positive effects on heterotrophic respiration and simultaneously increase evapotranspiration and lower the water table. This could increase aerobic decomposition and thus switch the Alaskan peatland from a C sink into a C source. Moreover, the increasing anthropogenic activities including land use will probably increase drought and subsequently enhance the risk of fire, releasing carbon to the atmosphere. The fate of Alaskan SOC stock and the biogeochemical cycling of the terrestrial ecosystems under future scenarios need further investigation.

Climate variables significantly affect C dynamics within each time slice. However, different vegetation distributions during various periods led to clear step changes, suggesting vegetation composition was likely to be another primary factor (Figs. 6, 7, 8, and 11). As key parameters controlling C dynamics in the model (e.g., maximum rate of photosynthesis, litter fall C, maximum rate of monthly NPP) are ecosystem type specific, vegetation distribution changes may drastically affect regional plant productivity and C storage. Our sensitivity test indicated that by replacing all vegetation types with forests, there was a total increase of 36.9 Pg in upland plus peatland soils. There was also an increase of 48.8 Pg C under warmer and wetter conditions, suggesting that both climate and vegetation distribution may have played important roles in carbon accumulation.

The vegetation changes reconstructed from fossil pollen data during different time periods followed the general climate history during the last 15 000 years. For instance, the migration of dark boreal forests over snow-covered tundra during the HTM was probably induced by the warmer and wetter climate that resulted from the insolation changes (He et al., 2014). The cooler and drier climate after the mid-Holocene limited the growth of boreal broadleaf conifers (Prentice et al., 1992), and therefore resulted in the replacement of broadleaf forest with needleleaf forest and tundra ecosystems. Since the parameters of our model for individual vegetation type were static, parameterizing the model using modern site-level observations might have introduced uncertainty to parameters, which may result in regional simulation uncertainties. Assuming each parameter as constant (e.g., the lowest water-table boundary; see Wang et al., 2016, for details) over time may also weaken the model's response to different climate scenarios. Furthermore, applying static vegetation maps at millennial scales and using modern elevation and pH data may simplify the complicated changes in landscape and terrestrial ecosystems, as vegetation can shift within hundreds of years (Ager and Brubake, 1985; see He et al., 2014, discussion section). Relatively coarse spatial resolution (0.5∘×0.5∘) in P-TEM simulations may also introduce uncertainties. In addition, because we used the modern inundation map to delineate the peatland and upland within each grid cell, we might have overestimated the total peatland area since not all inundated areas are peatlands. Linking field-estimated basal ages of peat cores to the vegetation types during each period involves large uncertainties due to the limitation of the peat classification and insufficient peat samples. Thus, the estimated spatially explicit basal age data shall also introduce a large uncertainty to our regional quantification of carbon accumulation.